Infrastructure tools to support an effective Radiation Oncology Learning Health System.

PURPOSE: Radiation Oncology Learning Health System (RO-LHS) is a promising approach to improve the quality of care by integrating clinical, dosimetry, treatment delivery, research data in real-time. This paper describes a novel set of tools to support the development of a RO-LHS and the current challenges they can address. METHODS: We present a knowledge graph-based approach to map radiotherapy data from clinical databases to an ontology-based data repository using FAIR concepts. This strategy ensures that the data are easily discoverable, accessible, and can be used by other clinical decision support systems. It allows for visualization, presentation, and data analyses of valuable information to identify trends and patterns in patient outcomes. We designed a search engine that utilizes ontology-based keyword searching, synonym-based term matching that leverages the hierarchical nature of ontologies to retrieve patient records based on parent and children classes, connects to the Bioportal database for relevant clinical attributes retrieval. To identify similar patients, a method involving text corpus creation and vector embedding models (Word2Vec, Doc2Vec, GloVe, and FastText) are employed, using cosine similarity and distance metrics. RESULTS: The data pipeline and tool were tested with 1660 patient clinical and dosimetry records resulting in 504 180 RDF (Resource Description Framework) tuples and visualized data relationships using graph-based representations. Patient similarity analysis using embedding models showed that the Word2Vec model had the highest mean cosine similarity, while the GloVe model exhibited more compact embeddings with lower Euclidean and Manhattan distances. CONCLUSIONS: The framework and tools described support the development of a RO-LHS. By integrating diverse data sources and facilitating data discovery and analysis, they contribute to continuous learning and improvement in patient care. The tools enhance the quality of care by enabling the identification of cohorts, clinical decision support, and the development of clinical studies and machine learning programs in radiation oncology.

3D deep convolution neural network for radiation pneumonitis prediction following stereotactic body radiotherapy.

In this study, we investigated 3D convolutional neural networks (CNNs) with input from radiographic and dosimetric datasets of primary lung tumors and surrounding lung volumes to predict the likelihood of radiation pneumonitis (RP). Pre-treatment, 3- and 6-month follow-up computed tomography (CT) and 3D dose datasets from one hundred and ninety-three NSCLC patients treated with stereotactic body radiotherapy (SBRT) were retrospectively collected and analyzed for this study. DenseNet-121 and ResNet-50 models were selected for this study as they are deep neural networks and have been proven to have high accuracy for complex image classification tasks. Both were modified with 3D convolution and max pooling layers to accept 3D datasets. We used a minority class oversampling approach and data augmentation to address the challenges of data imbalance and data scarcity. We built two sets of models for classification of three (No RP, Grade 1 RP, Grade 2 RP) and two (No RP, Yes RP) classes as outputs. The 3D DenseNet-121 models performed better (F1 score [0.81], AUC [0.91] [three class]; F1 score [0.77], AUC [0.84] [two class]) than the 3D ResNet-50 models (F1 score [0.54], AUC [0.72] [three-class]; F1 score [0.68], AUC [0.71] [two-class]) (p = 0.017 for three class predictions). We also attempted to identify salient regions within the input 3D image dataset via integrated gradient (IG) techniques to assess the relevance of the tumor surrounding volume for RP stratification. These techniques appeared to indicate the significance of the tumor and surrounding regions in the prediction of RP. Overall, 3D CNNs performed well to predict clinical RP in our cohort based on the provided image sets and radiotherapy dose information.

Automated data abstraction for quality surveillance and outcome assessment in radiation oncology.

Rigorous radiotherapy quality surveillance and comprehensive outcome assessment require electronic capture and automatic abstraction of clinical, radiation treatment planning, and delivery data. We present the design and implementation framework of an integrated data abstraction, aggregation, and storage, curation, and analytics software: the Health Information Gateway and Exchange (HINGE), which collates data for cancer patients receiving radiotherapy. The HINGE software abstracts structured DICOM-RT data from the treatment planning system (TPS), treatment data from the treatment management system (TMS), and clinical data from the electronic health records (EHRs). HINGE software has disease site-specific "Smart" templates that facilitate the entry of relevant clinical information by physicians and clinical staff in a discrete manner as part of the routine clinical documentation. Radiotherapy data abstracted from these disparate sources and the smart templates are processed for quality and outcome assessment. The predictive data analyses are done on using well-defined clinical and dosimetry quality measures defined by disease site experts in radiation oncology. HINGE application software connects seamlessly to the local IT/medical infrastructure via interfaces and cloud services and performs data extraction and aggregation functions without human intervention. It provides tools to assess variations in radiation oncology practices and outcomes and determines gaps in radiotherapy quality delivered by each provider.

Multi-View Data Integration Methods for Radiotherapy Structure Name Standardization.

Standardization of radiotherapy structure names is essential for developing data-driven personalized radiotherapy treatment plans. Different types of data are associated with radiotherapy structures, such as the physician-given text labels, geometric (image) data, and Dose-Volume Histograms (DVH). Prior work on structure name standardization used just one type of data. We present novel approaches to integrate complementary types (views) of structure data to build better-performing machine learning models. We present two methods, namely (a) intermediate integration and (b) late integration, to combine physician-given textual structure name features and geometric information of structures. The dataset consisted of 709 prostate cancer and 752 lung cancer patients across 40 radiotherapy centers administered by the U.S. Veterans Health Administration (VA) and the Department of Radiation Oncology, Virginia Commonwealth University (VCU). We used randomly selected data from 30 centers for training and ten centers for testing. We also used the VCU data for testing. We observed that the intermediate integration approach outperformed the models with a single view of the dataset, while late integration showed comparable performance with single-view results. Thus, we demonstrate that combining different views (types of data) helps build better models for structure name standardization to enable big data analytics in radiation oncology.

Machine-Learning Models for Multicenter Prostate Cancer Treatment Plans.

Clinical factors, including T-stage, Gleason score, and baseline prostate-specific antigen, are used to stratify patients with prostate cancer (PCa) into risk groups. This provides prognostic information for a heterogeneous disease such as PCa and guides treatment selection. In this article, we hypothesize that nonclinical factors may also impact treatment selection and their adherence to treatment guidelines. A total of 552 patients with intermediate- and high-risk PCa treated with definitive radiation with or without androgen deprivation therapy (ADT) between 2010 and 2017 were identified from 34 medical centers within the Veterans Health Administration. Medical charts were manually reviewed, and details regarding each patient's clinical history and treatment were extracted. Support Vector Machine and Random forest-based classification was used to identify clinical and nonclinical predictors of adherence to the treatment guidelines from the National Comprehensive Cancer Network (NCCN). We created models for predicting both initial treatment intent and treatment alterations. Our results demonstrate that besides clinical factors, the center in which the patient was treated (nonclinical factor) played a significant role in adherence to NCCN guidelines. Furthermore, the treatment center served as an important predictor to decide on whether or not to prescribe ADT; however, it was not associated with ADT duration and weakly associated with treatment alterations. Such center-bias motivates further investigation on details of center-specific barriers to both NCCN guideline adherence and on oncological outcomes. In addition, we demonstrate that publicly available data sets, for example, that from Surveillance, Epidemiology, and End Results (SEERs), may not be well equipped to build such predictive models on treatment plans.

Automatic Incident Triage in Radiation Oncology Incident Learning System.

The Radiotherapy Incident Reporting and Analysis System (RIRAS) receives incident reports from Radiation Oncology facilities across the US Veterans Health Affairs (VHA) enterprise and Virginia Commonwealth University (VCU). In this work, we propose a computational pipeline for analysis of radiation oncology incident reports. Our pipeline uses machine learning (ML) and natural language processing (NLP) based methods to predict the severity of the incidents reported in the RIRAS platform using the textual description of the reported incidents. These incidents in RIRAS are reviewed by a radiation oncology subject matter expert (SME), who initially triages some incidents based on the salient elements in the incident report. To automate the triage process, we used the data from the VHA treatment centers and the VCU radiation oncology department. We used NLP combined with traditional ML algorithms, including support vector machine (SVM) with linear kernel, and compared it against the transfer learning approach with the universal language model fine-tuning (ULMFiT) algorithm. In RIRAS, severities are divided into four categories; A, B, C, and D, with A being the most severe to D being the least. In this work, we built models to predict High (A & B) vs. Low (C & D) severity instead of all the four categories. Models were evaluated with macro-averaged precision, recall, and F1-Score. The Traditional ML machine learning (SVM-linear) approach did well on the VHA dataset with 0.78 F1-Score but performed poorly on the VCU dataset with 0.5 F1-Score. The transfer learning approach did well on both datasets with 0.81 F1-Score on VHA dataset and 0.68 F1-Score on the VCU dataset. Overall, our methods show promise in automating the triage and severity determination process from radiotherapy incident reports.

VA-Radiation Oncology Quality Surveillance Program.

PURPOSE: We sought to develop a quality surveillance program for approximately 15,000 US veterans treated at the 40 radiation oncology facilities at the Veterans Affairs (VA) hospitals each year. METHODS AND MATERIALS: State-of-the-art technologies were used with the goal to improve clinical outcomes while providing the best possible care to veterans. To measure quality of care and service rendered to veterans, the Veterans Health Administration established the VA Radiation Oncology Quality Surveillance program. The program carries forward the American College of Radiology Quality Research in Radiation Oncology project methodology of assessing the wide variation in practice pattern and quality of care in radiation therapy by developing clinical quality measures (QM) used as quality indices. These QM data provide feedback to physicians by identifying areas for improvement in the process of care and identifying the adoption of evidence-based recommendations for radiation therapy. RESULTS: Disease-site expert panels organized by the American Society for Radiation Oncology (ASTRO) defined quality measures and established scoring criteria for prostate cancer (intermediate and high risk), non-small cell lung cancer (IIIA/B stage), and small cell lung cancer (limited stage) case presentations. Data elements for 1567 patients from the 40 VA radiation oncology practices were abstracted from the electronic medical records and treatment management and planning systems. Overall, the 1567 assessed cases passed 82.4% of all QM. Pass rates for QM for the 773 lung and 794 prostate cases were 78.0% and 87.2%, respectively. Marked variations, however, were noted in the pass rates for QM when tumor site, clinical pathway, or performing centers were separately examined. CONCLUSIONS: The peer-review protected VA-Radiation Oncology Surveillance program based on clinical quality measures allows providers to compare their clinical practice to peers and to make meaningful adjustments in their personal patterns of care unobtrusively.

Effect of variations in atelectasis on tumor displacement during radiation therapy for locally advanced lung cancer.

PURPOSE: Atelectasis (AT), or collapsed lung, is frequently associated with central lung tumors. We investigated the variation of atelectasis volumes during radiation therapy and analyzed the effect of AT volume changes on the reproducibility of the primary tumor (PT) position. METHODS AND MATERIALS: Twelve patients with lung cancer who had AT and 10 patients without AT underwent repeated 4-dimensional fan beam computed tomography (CT) scans during radiation therapy per protocols that were approved by the institutional review board. Interfraction volume changes of AT and PT were correlated with PT displacements relative to bony anatomy using both a bounding box (BB) method and change in center of mass (COM). Linear regression modeling was used to determine whether PT and AT volume changes were independently associated with PT displacement. PT displacement was compared between patients with and without AT. RESULTS: The mean initial AT volume on the planning CT was 189 cm3 (37-513 cm3), and the mean PT volume was 93 cm3 (12-176 cm3). During radiation therapy, AT and PT volumes decreased on average 136.7 cm3 (20-369 cm3) for AT and 40 cm3 (-7 to 131 cm3) for PT. Eighty-three percent of patients with AT had at least one unidirectional PT shift that was greater than 0.5 cm outside of the initial BB during treatment. In patients with AT, the maximum PT COM shift was ≥0.5 cm in all patients and >1 cm in 58% of patients (0.5-2.4 cm). Changes in PT and AT volumes were independently associated with PT displacement (P < .01), and the correlation was smaller with COM (R2 = 0.58) compared with the BB method (R2 = 0.80). The median root mean squared PT displacement with the BB method was significantly less for patients without AT (0.45 cm) compared with those with AT (0.8cm, P = .002). CONCLUSIONS: Changes in AT and PT volumes during radiation treatment were significantly associated with PT displacements that often exceeded standard setup margins. Repeated 3-dimensional imaging is recommended in patients with AT to evaluate for PT displacements during treatment.

A longitudinal four-dimensional computed tomography and cone beam computed tomography dataset for image-guided radiation therapy research in lung cancer.

PURPOSE: To describe in detail a dataset consisting of serial four-dimensional computed tomography (4DCT) and 4D cone beam CT (4DCBCT) images acquired during chemoradiotherapy of 20 locally advanced, nonsmall cell lung cancer patients we have collected at our institution and shared publicly with the research community. ACQUISITION AND VALIDATION METHODS: As part of an NCI-sponsored research study 82 4DCT and 507 4DCBCT images were acquired in a population of 20 locally advanced nonsmall cell lung cancer patients undergoing radiation therapy. All subjects underwent concurrent radiochemotherapy to a total dose of 59.4-70.2 Gy using daily 1.8 or 2 Gy fractions. Audio-visual biofeedback was used to minimize breathing irregularity during all fractions, including acquisition of all 4DCT and 4DCBCT acquisitions in all subjects. Target, organs at risk, and implanted fiducial markers were delineated by a physician in the 4DCT images. Image coordinate system origins between 4DCT and 4DCBCT were manipulated in such a way that the images can be used to simulate initial patient setup in the treatment position. 4DCT images were acquired on a 16-slice helical CT simulator with 10 breathing phases and 3 mm slice thickness during simulation. In 13 of the 20 subjects, 4DCTs were also acquired on the same scanner weekly during therapy. Every day, 4DCBCT images were acquired on a commercial onboard CBCT scanner. An optically tracked external surrogate was synchronized with CBCT acquisition so that each CBCT projection was time stamped with the surrogate respiratory signal through in-house software and hardware tools. Approximately 2500 projections were acquired over a period of 8-10 minutes in half-fan mode with the half bow-tie filter. Using the external surrogate, the CBCT projections were sorted into 10 breathing phases and reconstructed with an in-house FDK reconstruction algorithm. Errors in respiration sorting, reconstruction, and acquisition were carefully identified and corrected. DATA FORMAT AND USAGE NOTES: 4DCT and 4DCBCT images are available in DICOM format and structures through DICOM-RT RTSTRUCT format. All data are stored in the Cancer Imaging Archive (TCIA, http://www.cancerimagingarchive.net/) as collection 4D-Lung and are publicly available. DISCUSSION: Due to high temporal frequency sampling, redundant (4DCT and 4DCBCT) data at similar timepoints, oversampled 4DCBCT, and fiducial markers, this dataset can support studies in image-guided and image-guided adaptive radiotherapy, assessment of 4D voxel trajectory variability, and development and validation of new tools for image registration and motion management.

A Voxel-by-Voxel Comparison of Deformable Vector Fields Obtained by Three Deformable Image Registration Algorithms Applied to 4DCT Lung Studies.

BACKGROUND: Commonly used methods of assessing the accuracy of deformable image registration (DIR) rely on image segmentation or landmark selection. These methods are very labor intensive and thus limited to relatively small number of image pairs. The direct voxel-by-voxel comparison can be automated to examine fluctuations in DIR quality on a long series of image pairs. METHODS: A voxel-by-voxel comparison of three DIR algorithms applied to lung patients is presented. Registrations are compared by comparing volume histograms formed both with individual DIR maps and with a voxel-by-voxel subtraction of the two maps. When two DIR maps agree one concludes that both maps are interchangeable in treatment planning applications, though one cannot conclude that either one agrees with the ground truth. If two DIR maps significantly disagree one concludes that at least one of the maps deviates from the ground truth. We use the method to compare 3 DIR algorithms applied to peak inhale-peak exhale registrations of 4DFBCT data obtained from 13 patients. RESULTS: All three algorithms appear to be nearly equivalent when compared using DICE similarity coefficients. A comparison based on Jacobian volume histograms shows that all three algorithms measure changes in total volume of the lungs with reasonable accuracy, but show large differences in the variance of Jacobian distribution on contoured structures. Analysis of voxel-by-voxel subtraction of DIR maps shows differences between algorithms that exceed a centimeter for some registrations. CONCLUSION: Deformation maps produced by DIR algorithms must be treated as mathematical approximations of physical tissue deformation that are not self-consistent and may thus be useful only in applications for which they have been specifically validated. The three algorithms tested in this work perform fairly robustly for the task of contour propagation, but produce potentially unreliable results for the task of DVH accumulation or measurement of local volume change. Performance of DIR algorithms varies significantly from one image pair to the next hence validation efforts, which are exhaustive but performed on a small number of image pairs may not reflect the performance of the same algorithm in practical clinical situations. Such efforts should be supplemented by validation based on a longer series of images of clinical quality.

Evaluation of 4-dimensional computed tomography to 4-dimensional cone-beam computed tomography deformable image registration for lung cancer adaptive radiation therapy.

PURPOSE: To evaluate 2 deformable image registration (DIR) algorithms for the purpose of contour mapping to support image-guided adaptive radiation therapy with 4-dimensional cone-beam CT (4DCBCT). METHODS AND MATERIALS: One planning 4D fan-beam CT (4DFBCT) and 7 weekly 4DCBCT scans were acquired for 10 locally advanced non-small cell lung cancer patients. The gross tumor volume was delineated by a physician in all 4D images. End-of-inspiration phase planning 4DFBCT was registered to the corresponding phase in weekly 4DCBCT images for day-to-day registrations. For phase-to-phase registration, the end-of-inspiration phase from each 4D image was registered to the end-of-expiration phase. Two DIR algorithms-small deformation inverse consistent linear elastic (SICLE) and Insight Toolkit diffeomorphic demons (DEMONS)-were evaluated. Physician-delineated contours were compared with the warped contours by using the Dice similarity coefficient (DSC), average symmetric distance, and false-positive and false-negative indices. The DIR results are compared with rigid registration of tumor. RESULTS: For day-to-day registrations, the mean DSC was 0.75 ± 0.09 with SICLE, 0.70 ± 0.12 with DEMONS, 0.66 ± 0.12 with rigid-tumor registration, and 0.60 ± 0.14 with rigid-bone registration. Results were comparable to intraobserver variability calculated from phase-to-phase registrations as well as measured interobserver variation for 1 patient. SICLE and DEMONS, when compared with rigid-bone (4.1 mm) and rigid-tumor (3.6 mm) registration, respectively reduced the average symmetric distance to 2.6 and 3.3 mm. On average, SICLE and DEMONS increased the DSC to 0.80 and 0.79, respectively, compared with rigid-tumor (0.78) registrations for 4DCBCT phase-to-phase registrations. CONCLUSIONS: Deformable image registration achieved comparable accuracy to reported interobserver delineation variability and higher accuracy than rigid-tumor registration. Deformable image registration performance varied with the algorithm and the patient.

Dose escalation for locally advanced lung cancer using adaptive radiation therapy with simultaneous integrated volume-adapted boost.

PURPOSE: To test the feasibility of a planned phase 1 study of image-guided adaptive radiation therapy in locally advanced lung cancer. METHODS AND MATERIALS: Weekly 4-dimensional fan beam computed tomographs (4D FBCT) of 10 lung cancer patients undergoing concurrent chemoradiation therapy were used to simulate adaptive radiation therapy: After an initial intensity modulated radiation therapy plan (0-30 Gy/2 Gy), adaptive replanning was performed on week 2 (30-50 Gy/2 Gy) and week 4 scans (50-66 Gy/2 Gy) to adjust for volume and shape changes of primary tumors and lymph nodes. Week 2 and 4 clinical target volumes (CTV) were deformably warped from the initial planning scan to adjust for anatomical changes. On the week 4 scan, a simultaneous integrated volume-adapted boost was created to the shrunken primary tumor with dose increases in 5 0.4-Gy steps from 66 Gy to 82 Gy in 2 scenarios: plan A, lung isotoxicity; plan B, normal tissue tolerance. Cumulative dose was assessed by deformably mapping and accumulating biologically equivalent dose normalized to 2 Gy-fractions (EQD2). RESULTS: The 82-Gy level was achieved in 1 in 10 patients in scenario A, resulting in a 13.4-Gy EQD2 increase and a 22.1% increase in tumor control probability (TCP) compared to the 66-Gy plan. In scenario B, 2 patients reached the 82-Gy level with a 13.9 Gy EQD2 and 23.4% TCP increase. CONCLUSIONS: The tested image-guided adaptive radiation therapy strategy enabled relevant increases in EQD2 and TCP. Normal tissue was often dose limiting, indicating a need to modify the present study design before clinical implementation.

A protocol to extend the longitudinal coverage of on-board cone-beam CT.

The longitudinal coverage of a LINAC-mounted CBCT scan is limited to the corresponding dimensional limits of its flat panel detector, which is often shorter than the length of the treatment field. These limits become apparent when fields are designed to encompass wide regions, as when providing nodal coverage. Therefore, we developed a novel protocol to acquire double orbit CBCT images using a commercial system, and combine the images to extend the longitudinal coverage for image-guided adaptive radiotherapy (IGART). The protocol acquires two CBCT scans with a couch shift similar to the "step-and-shoot" cine CT acquisition, allowing a small longitudinal overlap of the two reconstructed volumes. An in-house DICOM reading/writing software was developed to combine the two image sets into one. Three different approaches were explored to handle the possible misalignment between the two image subsets: simple stacking, averaging the overlapped volumes, and a 3D-3D image registration with the three translational degrees of freedom. Using thermoluminescent dosimeters and custom-designed holders for a CTDI phantom set, dose measurements were carried out to assess the resultant imaging dose of the technique and its geometric distribution. Deformable registration was tested on patient images generated with the double-orbit protocol, using both the planning FBCT and the artificially deformed CBCT as source images. The protocol was validated on phantoms and has been employed clinically for IRB-approved IGART studies for head and neck and prostate cancer patients.

Classifying geometric variability by dominant eigenmodes of deformation in regressing tumours during active breath-hold lung cancer radiotherapy.

The purpose of this study is to develop and evaluate a lung tumour interfraction geometric variability classification scheme as a means to guide adaptive radiotherapy and improve measurement of treatment response. Principal component analysis (PCA) was used to generate statistical shape models of the gross tumour volume (GTV) for 12 patients with weekly breath hold CT scans. Each eigenmode of the PCA model was classified as 'trending' or 'non-trending' depending on whether its contribution to the overall GTV variability included a time trend over the treatment course. Trending eigenmodes were used to reconstruct the original semi-automatically delineated GTVs into a reduced model containing only time trends. Reduced models were compared to the original GTVs by analyzing the reconstruction error in the GTV and position. Both retrospective (all weekly images) and prospective (only the first four weekly images) were evaluated. The average volume difference from the original GTV was 4.3% ± 2.4% for the trending model. The positional variability of the GTV over the treatment course, as measured by the standard deviation of the GTV centroid, was 1.9 ± 1.4 mm for the original GTVs, which was reduced to 1.2 ± 0.6 mm for the trending-only model. In 3/13 cases, the dominant eigenmode changed class between the prospective and retrospective models. The trending-only model preserved GTV and shape relative to the original GTVs, while reducing spurious positional variability. The classification scheme appears feasible for separating types of geometric variability by time trend.

Optimizing principal component models for representing interfraction variation in lung cancer radiotherapy.

PURPOSE: To optimize modeling of interfractional anatomical variation during active breath-hold radiotherapy in lung cancer using principal component analysis (PCA). METHODS: In 12 patients analyzed, weekly CT sessions consisting of three repeat intrafraction scans were acquired with active breathing control at the end of normal inspiration. The gross tumor volume (GTV) and lungs were delineated and reviewed on the first week image by physicians and propagated to all other images using deformable image registration. PCA was used to model the target and lung variability during treatment. Four PCA models were generated for each specific patient: (1) Individual models for the GTV and each lung from one image per week (week to week, W2W); (2) a W2W composite model of all structures; (3) individual models using all images (weekly plus repeat intrafraction images, allscans); and (4) composite model with all images. Models were reconstructed retrospectively (using all available images acquired) and prospectively (using only data acquired up to a time point during treatment). Dominant modes representing at least 95% of the total variability were used to reconstruct the observed anatomy. Residual reconstruction error between the model-reconstructed and observed anatomy was calculated to compare the accuracy of the models. RESULTS: An average of 3.4 and 4.9 modes was required for the allscans models, for the GTV and composite models, respectively. The W2W model required one less mode in 40% of the patients. For the retrospective composite W2W model, the average reconstruction error was 0.7 +/- 0.2 mm, which increased to 1.1 +/- 0.5 mm when the allscans model was used. Individual and composite models did not have significantly different errors (p = 0.15, paired t-test). The average reconstruction error for the prospective models of the GTV stabilized after four measurements at 1.2 +/- 0.5 mm and for the composite model after five measurements at 0.8 +/- 0.4 mm. CONCLUSIONS: Retrospective PCA models were capable of reconstructing original GTV and lung shapes and positions within several millimeters with three to four dominant modes, on average. Prospective models achieved similar accuracy after four to five measurements.

Clinical evaluation of soft tissue organ boundary visualization on cone-beam computed tomographic imaging.

PURPOSE: Cone-beam computed tomographic images (CBCTs) are increasingly used for setup correction, soft tissue targeting, and image-guided adaptive radiotherapy. However, CBCT image quality is limited by low contrast and imaging artifacts. This analysis investigates the detectability of soft tissue boundaries in CBCT by performing a multiple-observer segmentation study. METHODS AND MATERIALS: In four prostate cancer patients prostate, bladder and rectum were repeatedly delineated by five observers on CBCTs and fan-beam CTs (FBCTs). A volumetric analysis of contouring variations was performed by calculating coefficients of variation (COV: standard deviation/average volume). The topographical distribution of contouring variations was analyzed using an average surface mesh-based method. RESULTS: Observer- and patient-averaged COVs for FBCT/CBCT were 0.09/0.19 for prostate, 0.05/0.08 for bladder, and 0.09/0.08 for rectum. Contouring variations on FBCT were significantly smaller than on CBCT for prostate (p < 0.03) and bladder (p < 0.04), but not for rectum (p < 0.37; intermodality differences). Intraobserver variations from repeated contouring of the same image set were not significant for either FBCT or CBCT (p < 0.05). Average standard deviations of individual observers' contour differences from average surface meshes on FBCT vs. CBCT were 1.5 vs. 2.1 mm for prostate, 0.7 vs. 1.4 mm for bladder, and 1.3 vs. 1.5 mm for rectum. The topographical distribution of contouring variations was similar for FBCT and CBCT. CONCLUSION: Contouring variations were larger on CBCT than FBCT, except for rectum. Given the well-documented uncertainty in soft tissue contouring in the pelvis, improvement of CBCT image quality and establishment of well-defined soft tissue identification rules are desirable for image-guided radiotherapy.

Development of a population-based model of surface segmentation uncertainties for uncertainty-weighted deformable image registrations.

PURPOSE: To develop a population-based model of surface segmentation uncertainties for uncertainty-weighted surface-based deformable registrations. METHODS: The contours of the prostate, the bladder, and the rectum were manually delineated by five observers on fan beam CT images of four prostate cancer patients. First, patient-specific representations of structure segmentation uncertainties were derived by determining the interobserver variability (i.e., standard deviation) of the structure boundary delineation. This was achieved by (1) generating an average structure surface mesh from the structure contours drawn by different observers, and (2) calculating three-dimensional standard deviation surface meshes (SDSMs) based on the perpendicular distances from the individual boundary surface meshes to the average surface mesh computed above. Then an average structure surface mesh was constructed to be the reference mesh for the population-based model. The average structure meshes of the other patients were deformably registered to the reference mesh. The calculated deformable vector fields were used to map the patient-specific SDSMs to the reference mesh to obtain the registered SDSMs. Finally, the population-based SDSM was derived by taking the average of the registered SDSMs in quadrature. RESULTS: Population-based structure surface statistical models of the prostate, the bladder, and the rectum were created by mapping the patient-specific SDSMs to the population surface model. Graphical visualization indicates that the boundary uncertainties are dependent on anatomical location. CONCLUSIONS: The authors have developed and demonstrated a general method for objectively constructing surface maps of uncertainties derived from topologically complex structure boundary segmentations from multiple observers. The computed boundary uncertainties have significant spatial variations. They can be used as weighting factors for surface-based probabilistic deformable registration.